Date: November 1, 2018
Author(s): Rob Williams
For those who need a workstation graphics card that should slaughter benchmarks, not wallets, AMD’s Radeon Pro WX 8200 is worth checking out. This Vega-based WX sports a ton of cores for great compute and graphics performance, as well as 8GB of ECC HBM2 memory. But how does it serve your workload? Let’s figure that out.
I posted an in-depth look at AMD’s Radeon ProRender renderer last week, and in that article, I mentioned that the workstation side of Radeon has been on fire as of late. New ProRender versions are coming out left and right, and in August, we were even greeted to a brand-new Radeon Pro GPU: the WX 8200.
With the name “WX 8200”, the card is a bit of an oddball given the (X)100 naming seen before it. With the 8200, AMD effectively introduces the second release cycle for Radeon Pro, though I’m not sure it’s safe to call it “second gen”. Nor would I want to call it that – the WX 8200 is similar in hardware to the RX Vega 56, and the WX 9100 is like a Vega 64 – both are of the same generation.
Nonetheless, the WX 8200 offers a lot for its price-point in the workstation market. It’s capped at 8GB memory, which might throw some off, but it’s at least ECC (error correction), which is not something we see at the $1,000 price-point often (or ever?). With 3,584 Vega cores under-the-hood, the WX 8200 is no slouch when it comes to raw performance, either for compute, or gaming.
The only cards in AMD’s Radeon Pro lineup to beat out the WX 8200 in performance would be the WX 9100 and SSG. Both of those cards have 4,096 cores, and a full 16GB of memory. The SSG is special in that it has 2TB of SSD storage on the card itself (four drives in RAID 0), allowing the GPU to get even closer to the project cache (eg: with Adobe Premiere Pro), in effect removing a middleman in the transfer.
|AMD’s Radeon Pro Workstation GPU Lineup|
|Cores||Base MHz||Peak FP32||Memory||Bandwidth||TDP||Price|
|SSG 9||4096||1440||12.3 TFLOPS||16 GB 8||484 GB/s||260W||$6999|
|WX 9100||4096||1200||12.3 TFLOPS||16 GB 8||484 GB/s||230W||$1399|
|WX 8200||3584||1200||10.8 TFLOPS||8 GB 8||512 GB/s||230W||$999|
|Frontier||4096||1382||13.1 TFLOPS||16 GB 4||484 GB/s||300W||$499|
|Pro Duo||2304 x2||1243||5.7 TFLOPS||32 GB 3||448 GB/s||250W||$449|
|WX 7100||2304||1188||5.73 TFLOPS||8 GB 3||224 GB/s||130W||$549|
|WX 5100||1792||713||3.89 TFLOPS||8 GB 3||160 GB/s||75W||$359|
|WX 4100||1024||1125||2.46 TFLOPS||4 GB 3||96 GB/s||50W||$259|
|WX 3100||512||925||1.25 TFLOPS||4 GB 3||96 GB/s||50W||$169|
|WX 2100||512||925||1.25 TFLOPS||2 GB 3||56 GB/s||50W||$129|
|Notes||1 GDDR6; 2 GDDR5X; 3 GDDR5; 4 HBM2
5 GDDR6 (ECC); 6 GDDR5X (ECC); 7 GDDR5 (ECC); 8 HBM2 (ECC)
9 Includes 2TB of solid-state storage on-card.
Architecture: WX 2100~7100 = Polaris; WX 8200, 9100 & SSG = Vega
Tested GPUs are italicized
The WX 8200 is a great all-around card thanks to its $999 price-point and the amount of performance it offers. It’s not going to beat out the competition in all tests, but it performs extremely well in a few tests. For comparison’s sake, the performance results here also include the Vega 64, so we can easily see which tests benefit from RPro performance optimizations, and which don’t.
I remember when the WX 9100 launched at $2,199, but here we are, with the same card being sold regularly for around $1,399. Even before looking at the results, I’d have to wager that if I were choosing to go with a WX 8200, I’d think long and hard about whether or not I’d ever want more than 8GB of memory, and the extra performance. You don’t gain an enormous amount by stepping up to the WX 9100, but it’s what I’d consider the more future-proofed card.
On the following pages, the results of our WS GPU test gauntlet will be seen. The tests chosen cover a wide-range of scenarios, from rendering to compute, and includes the use of both synthetic benchmarks and tests with real-world applications from the likes of Adobe and Autodesk.
14 GPUs are being tested for this article, although because we had a second TITAN Xp on-hand, dual-GPU results will also appear throughout. If the dual-GPU configuration isn’t found in a chart, it means there was no performance scaling whatsoever across the multiple cards.
Here are the specs of the test machine:
|Techgage Workstation Test System|
|Processor||Intel Core i9-7980XE (18-core; 2.6GHz)|
|Motherboard||ASUS ROG STRIX X299-E GAMING|
|Memory||HyperX FURY (4x16GB; DDR4-2666 16-18-18)|
|Graphics||AMD Radeon RX Vega 64 8GB (18.10.1)
AMD Radeon RX 580 8GB (Red Devil; 18.10.1)
AMD Radeon Pro WX 8200 8GB (18.Q3.1)
AMD Radeon Pro WX 7100 8GB (18.Q3.1)
AMD Radeon Pro WX 5100 4GB (18.Q3.1)
AMD Radeon Pro WX 4100 4GB (18.Q3.1)
AMD Radeon Pro WX 3100 4GB (18.Q3.1)
NVIDIA GeForce RTX 2080 Ti 11GB (NVIDIA FE; 416.34)
NVIDIA TITAN Xp 12GB (416.34)
NVIDIA GeForce GTX 1080 Ti 11GB (416.34)
NVIDIA Quadro P6000 24GB (416.30)
NVIDIA Quadro P5000 16GB (416.30)
NVIDIA Quadro P4000 8GB (416.30)
NVIDIA Quadro P2000 5GB (416.30)
|Storage||Kingston KC1000 960GB M.2 SSD|
|Power Supply||Corsair 80 Plus Gold AX1200|
|Chassis||Corsair Carbide 600C Inverted Full-Tower|
|Cooling||NZXT Kraken X62 AIO Liquid Cooler|
|Et cetera||Windows 10 Pro build 17763 (1809)|
|For an in-depth pictorial look at this build, head here.|
Benchmark results are categorized and spread across the next four pages. On page 2, Adobe’s Premiere Pro and MAGIX’s Vegas Pro lead our encoding tests, with both AVC and HEVC codecs taken care of. On the same page, Sandra’s financial and scientific performance can be seen, as well as the cryptography.
On page 3, a few renderers are taken care of. These include the popular open-source design suite Blender, as well as LuxMark, and Radeon ProRender. ProRender is being tested with the help of 3ds Max, a suite that normally handles Redshift and V-Ray as well – but alas, neither work (or work well) on AMD graphics cards.
Page 4 is home to viewport performance, covered with the help of SPEC and its SPECviewperf suite, as well as Cadalyst, and its namesake AutoCAD test. In total, 9 test results are featured here, covering important design suites like CATIA, SolidWorks, Siemens NX, Creo, as well as Autodesk’s 3ds Max and Maya.
Finally, page 5 is home to a couple of gaming tests (more for sanity checks, since gaming isn’t going to be a huge focus here), as well as final thoughts.
So without further ado, let’s get this train moving.
With the straight-forward 4K to 1080p encode, all of the big cards perform essentially the same. When a real project is brought into the mix, the cards get separated up a lot more. Both the Vega 64 and WX 8200 perform the same, and NVIDIA’s Quadro P4000 beats out both in the Techgage encode.
With the 8K to 4K encode, it becomes easier to understand that you definitely want some decent horsepower if your goal is video work in Premiere Pro. The smaller cards struggle, making Quadro P4000+ or Radeon Pro WX 8200+ an ideal match – though a gaming card like the RX 580 does not fall that far behind, realistically.
We’ve seen strong performance from AMD in MAGIX’s Vegas Pro before, and that continues to be the case with the latest version – for the most part.
Interestingly, while I thought that the Taipei 101 project was the more grueling of the two here AMD’s cards struggled with the Air Canada one overall. There, the lowly Quadro P2000 managed to outperform a relative beast like the WX 8200. With the Taipei project, AMD’s strengths return.
Is it a surprise that AMD performs so well in cryptography? Here, the Vega 64 slides just in behind the RTX 2080 Ti, and ahead of the TITAN Xp. The WX 8200, meanwhile, jeers that same TITAN Xp from a short distance away. Fantastic performance overall.
AMD continues to perform great in the scientific test, sitting behind the Vega 64 and ahead of every GeForce and Radeon – save for the Turing-based RTX 2080 Ti. What matters most with this kind of performance is the speed of the GPU, but AMD clearly has certain optimizations for this workload that NVIDIA doesn’t. Or, it at least doesn’t have enough to redeem itself here. You know… ignoring the fact that the RTX 2080 Ti dominates here. That doesn’t explain the faster TITAN Xp falling behind the WX 8200, though.
With the financial test, AMD’s cards do not perform quite as strongly. Here, NVIDIA absolutely slaughters the entire collection with the RTX 2080 Ti performing better than dual TITAN Xps.
The financial score is an aggregate result of three algorithm tests, involving Black-Scholes, Binomial, and Monte Carlo. With Black-Scholes, the 2080 Ti performs beyond the TITAN Xp what you’d expect, but performance in Binomial and Monte Carlo is literally doubled, hence the domination here.
Recent AMD CPUs and GPUs alike love Blender’s Cycles renderer, and the best proof of that is seen near the top: the Radeon RX Vega 64 beat out every one of NVIDIA’s GPUs. That even includes the TITAN Xp. The WX 8200 is similar hardware to a Vega 56, and it places about where I’d expect. It sits real close to NVIDIA’s Pascal-based Quadro P6000.
As the TITAN Xp is the technically more powerful card of the two, AMD clearly has some great optimizations going on with this popular design tool. WX 8200 even beats out NVIDIA’s Quadro P5000, a GPU that costs about $1,700 at the current time (though it has 16GB of VRAM).
The WX 9100 (cue “if I had one!” meme) would overlap Vega 64’s positioning, giving AMD a strong lead in this suite overall.
There are a few things to talk about here. For starters, the results at the bottom are correct: the WX 3100 for some reason performed better in the LuxBall render over the WX 4100, despite the opposite being true for the Hotel render. Beyond that, everything else scales pretty much as expected.
The WX 8200 performs directly behind the TITAN Xp in the LuxBall render, and a bit behind the Vega 64 in the Hotel render, as we’d expect.
Unfortunately for AMD, NVIDIA’s new Turing RTX cards appear to be really good at LuxMark. Compared to the GTX 1080 Ti, the RTX 2080 Ti scores almost double the LuxBall score, and about 70% more in the Hotel score. It might be a good thing RTX doesn’t work in Blender right now, or we may have seen a different picture painted above.
I am only going to touch on ProRender performance briefly here, as I covered it last week in greater detail, including a look at CPU+GPU heterogeneous rendering.
Overall, NVIDIA rules ProRender, which is like saying AMD rules V-Ray (but that one’s not true). I had thought before that AMD’s Radeon Pros offered ProRender optimizations, but it’s clear that’s not the case, with the slightly faster clocked RX 580 outperforming the equally geared WX 7100.
In ProRender, the more GPU horsepower you have, the better. That might seem strange to say when the dual-GPU configuration in the chart performs worse than the single GPU complement, but, that’s because multi-GPU on NVIDIA is just not supported well in ProRender (at this time). As far as I understand it, dual Radeon Pro cards will have no problem scaling as you’d expect with this renderer.
Sheer strength is the name of the game with 3ds Max. Quadros and Radeon Pros don’t seem to offer special performance optimizations at all, a good thing for those who use the suite. Note that this wouldn’t necessarily mirror performance if you were use a plugin-specific viewport, but the faster the GPU you’re equipped with, the happier you’ll be editing in Max.
Maya shows very similar scaling to Max, except here, Radeon Pros do have specific performance optimizations. That helps the WX 8200 leap ahead of the technically faster RX Vega 64. It also places the card right behind the Quadro P5000, which is currently priced about seventy percent higher than AMD’s WX 8200 (but again, NVIDIA’s card has double the VRAM).
On AMD’s side, performance optimizations seem to be in place that helps push the WX 8200 to sit just ahead of the Vega 64 at 1080p, but just behind at 4K. Given this scaling, it seems likely that the WX 9100 would outperform the Vega 64, but it’d still sit behind the slightly lesser expensive Quadro P5000.
NVIDIA’s cards are a little interesting in that CATIA makes use of performance optimizations in the Quadros, but also the TITAN Xp. The GeForce cards, including the 1080 Ti and the newly released RTX 2080 Ti, both fall to the middle of the pack, behind the P4000, which is equivalent to a GTX 1060/1070.
We’ve seen good performance from the WX 8200 throughout most of the tests up to this point, but its SolidWorks performance is notable. Whereas AMD delivered similar performance on both Radeon and Radeon Pro in CATIA, the professional cards are blessed with specific performance optimizations in SolidWorks. Here, the WX 8200 sits just behind the Quadro P5000, and in third place overall, and outperforming the TITAN Xp too.
NVIDIA saves the best SolidWorks performance for its Quadro cards, which means the gaming cards are simply not a great choice. Even a modest Quadro P2000 outperforms the RTX 2080 Ti. Not by much, mind you, but it still happens – and you sure wouldn’t expect it.
I like to say “know your workload” a lot, and it’s with applications like Siemens’ NX that makes it really important. Gaming GPUs on both sides of the fence fall to the absolute bottom, though it’s notable that the Radeons double the performance of the GeForces (look at the RX 580 with double the performance of an RTX 2080 Ti, but at such low levels, it’s largely inconsequential). You can get dramatically improved performance by opting for even the smallest Quadro or Radeon Pro in the chart over any gaming card.
We’ve returned to a world where brute force matters more, not so much optimizations – at least where performance is concerned. That said, the RTX 2080 Ti doesn’t exactly slay the TITAN Xp as you’d expect it to, which means there could be room for growth here. The Qudros do slightly better than their gaming counterparts, but not by a huge margin.
This is another suite where the WX 8200 falls just behind the P5000, despite costing a lot less. The WX 8200 also performs a lot better than the Vega 64, again hinting at specific Radeon Pro optimizations, along with the WX 7100 ahead of the RX 580.
In both the medical and energy tests, the cards scale pretty much as you’d expect given their rated performance. I will say that these results only paint part of the overall picture, though. On the smaller Radeon Pros, the energy test took ages to get through in comparison to every other card. If you are involved in purchasing hardware for these industries, chances are good you know not to go with cards found closer to the bottom of the performance ladder.
That said, in both of these tests, NVIDIA outperforms AMD pretty easily overall, with 1080p performance being a lot better in the medical test – for some reason.
I left AutoCAD for the end as it’s the only test result here not gained through SPECviewperf. Instead, it’s tested with Cadalyst, a benchmark script that goes by the same name as the website that produced it.
Based on the results above, NVIDIA appears to offer the same AutoCAD performance optimizations on both its GeForce and Quadros, but AMD offers its own in our dreams. It’s not often we see such a clear divide between one vendor and another in a chart. For the best AutoCAD performance, you’re not likely going to be scouting out a Radeon Pro.
I’ll talk about the gaming performance in one fell swoop, because the scaling is almost identical across the board. Admittedly, I had wanted to include Unigine’s Superposition as a bit of a second opinion, but oddities crept into our automated testing, so I am not sure which results to trust.
Nonetheless, the WX 8200 performs behind the Vega 64 in all three tests to the same degree you’d expect to see the Vega 56 to. That means it settles in right behind NVIDIA’s Quadro P5000. Overall, not too much to comment on here, but good performance overall, even if it’s what we’d expect to see. I didn’t dive into real games this go around, but it goes without saying that the WX 8200 is more than capable of delivering a quality experience (great for 1440p or modest 4K).
When AMD unveiled its Radeon Pro WX 8200 at SIGGRAPH, it was a little confusing for two reasons. For starters, there was a “2” in the name. As mentioned earlier, this is simply to signify the second cycle of Radeon Pro devices. The second reason for confusion was the choice to ship with 8GB of VRAM. For the cost, many people expected more, and while I think it would be nice to see more, I don’t think it’s going to affect most people who are going to use the card – and if there is a limitation, that’s why it again pays to know your workload.
The fact of the matter is, HBM2 is expensive, and AMD couldn’t put 16GB of the stuff on this card for the same reason we don’t see 16GB on the Vega 64 – and likewise why we only see 16GB on the bigger WX 9100. AMD could have tacked on more memory if it wanted to charge more, and at that point, it’d fight even closer to the WX 9100, which would be strange (especially with that card available at $1,399).
It won’t matter if you don’t use it, but it’s important to note that the WX 8200 ships with ECC memory, which is something that surprises me a bit. The WX 7100 didn’t have any, and for NVIDIA, you need to go with the ~$1,700 Quadro P5000 to get some (though to be fair, that card does include 16GB, of GDDR5X).
Given what the Radeon Pro WX 8200 offers, I’d wager the $999 asking price is fair. From a gaming perspective, it has the power to deliver quality experiences, and the same can be said for design. As with any workstation GPU, you really need to understand where benefits can be seen, and on which vendor’s GPU. With Siemens NX, for example, you’d screw yourself over if you assumed gaming cards would be suitable enough. There, pro cards offer obvious benefits.
The WX 8200 doesn’t dominate most workloads, but it does deliver super-strong performance in some cases, such as with cryptography and scientific performance, as well as great performance in CATIA, SolidWorks, Siemens NX, and solid performance in Creo, Maya and 3ds Max.
As with any new workstation GPU that hits the lab, the WX 8200 isn’t just going to go sit on a shelf. It’ll be kicking around, being retested again and again over time, especially as other new cards land, and more tests get added to the suite.
As I mentioned in our news section a couple of weeks ago, I’ve been in the process of scripting both the installation and benchmarking process of our suites (or at least most of it). Currently, I’m making good headway, and hope to be finished in the next couple of weeks (somewhat throttled by upcoming travel). Long story short: once that chore is done, the next step will be enhancing the tests that exist, and augment the suite with other tests people want to see. That includes DaVinci Resolve and Capture One at the forefront, but others are in the midst to be explored as well.
If you have questions about the WX 8200 that I didn’t answer here, please leave a comment!
Copyright © 2005-2021 Techgage Networks Inc. - All Rights Reserved.